CN107044947B - A kind of recognition methods of the PM2.5 pollution index based on characteristics of image - Google Patents
A kind of recognition methods of the PM2.5 pollution index based on characteristics of image Download PDFInfo
- Publication number
- CN107044947B CN107044947B CN201710301867.4A CN201710301867A CN107044947B CN 107044947 B CN107044947 B CN 107044947B CN 201710301867 A CN201710301867 A CN 201710301867A CN 107044947 B CN107044947 B CN 107044947B
- Authority
- CN
- China
- Prior art keywords
- image
- value
- pixel
- weighted average
- queue
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 84
- 238000013528 artificial neural network Methods 0.000 claims description 31
- 238000004364 calculation method Methods 0.000 claims description 12
- 230000008859 change Effects 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 10
- 230000001537 neural effect Effects 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 210000004027 cell Anatomy 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 4
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 3
- 230000005540 biological transmission Effects 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 235000013399 edible fruits Nutrition 0.000 claims description 2
- 238000004321 preservation Methods 0.000 claims description 2
- 238000005259 measurement Methods 0.000 abstract description 7
- 230000000694 effects Effects 0.000 abstract description 2
- 239000012528 membrane Substances 0.000 description 16
- 239000013618 particulate matter Substances 0.000 description 8
- 238000005070 sampling Methods 0.000 description 7
- 238000012549 training Methods 0.000 description 5
- 230000010355 oscillation Effects 0.000 description 4
- 239000000126 substance Substances 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 3
- 239000010453 quartz Substances 0.000 description 3
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N silicon dioxide Inorganic materials O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 description 3
- 230000005250 beta ray Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 239000003153 chemical reaction reagent Substances 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000004062 sedimentation Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000010186 staining Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
- G01N15/06—Investigating concentration of particle suspensions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
- G01N15/06—Investigating concentration of particle suspensions
- G01N15/075—Investigating concentration of particle suspensions by optical means
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Pathology (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Immunology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Dispersion Chemistry (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to image procossing and meteorology crossing domains, more particularly to a kind of recognition methods of PM2.5 pollution index based on characteristics of image, the measuring method of existing PM2.5 is mainly solved there are the complicated for operation and inaccurate technical problem of measurement result, the history image for being labelled with PM2.5 measured value is uniformly scaled X by the present inventionS×YSPixel size is simultaneously converted to grayscale image;Its characteristics of image is calculated to each grayscale image;The prediction model of PM2.5 is obtained using suitable homing method using characteristics of image as input;Target image is all scaled XS×YSPixel size is simultaneously converted to grayscale image;Its various feature is calculated for the target image after scaling;Target image PM2.5 value is calculated using the characteristics of image of target image as the input for having trained prediction model.The present invention is analyzed to estimate the PM2.5 value of the described scene of image by the feature to piece image, so as to be applied to the mankind's activities such as daily trip, health, weather forecasting, and can be used for the rapid survey of extensive PM2.5 pollution index.
Description
Technical field
The invention belongs to image procossing and meteorology crossing domains, and in particular to a kind of PM2.5 based on characteristics of image is dirty
The recognition methods of staining index.
Background technique
Afloat particulate matter is of different sizes in air, and PM2.5 is wherein relatively fine part.Measure the dense of PM2.5
Degree is generally divided into two steps: (1) PM2.5 and biggish particulate separation, almost all of measuring method requires this step;
(2) weight for the PM2.5 that measurement is separated.Currently, there are three types of the PM2.5 measuring methods that environmental administration of various countries are widely used:
Gravimetric method, β attenuation sensors and trace oscillating balance method.
PM2.5 manual monitoring method is also known as standard weight method (filter membrane weight method) and belongs to manual monitoring method, is mainly used for
Research, or the calibration to automatic analysis method.This method is easier, it is only necessary to a PM2.5 cutting head, a pump and film frame
And its filter membrane, after acquiring 24 hours samples, remove filter membrane weighing, it may be necessary to 3 samples of parallel acquisition, through constant temperature perseverance
It is re-weighed after wet.
The advantages of filter membrane weight method is that economic cost is low, is easy to implement.Disadvantage: (1) air-flow constantly passes through sampling for a long time
Filter membrane, collected substance will also result in the damage of volatility and half volatile substance with the variation of air-flow and temperature on filter membrane
It loses;(2) some superfine small particles pass through filter membrane, cause Lower result;(3) gaseous material may be adsorbed by filter membrane, cause to tie
Fruit is higher.
The survey station point of automatic monitoring method is unattended, and data are directly obtained by network, including β attenuation sensors and micro-
Measure oscillating balance method.The basic principle of β attenuation sensors is the β discharged using the particulate matter being deposited on quartz filter to C14
The variation of the variation detection airborne particulate amount of substance of ray attenuation amount.Surrounding air sucks sampling pipe through cutter by sampling pump,
Go out through filter membrane heel row.Particulate matter is deposited on strip quartz filter, and when β ray passes through the filter membrane for depositing particulate matter, β is penetrated
Line intensity decays, and the concentration of particulate matter is calculated by the tester to attenuation.Micro oscillating balance (TEOM) method is also known as
Micro quartz oscillation sedimentation balance method, this method are using a quartzy hollow conical pipe in mass sensor, in hollow conical pipe
Replaceable filter membrane is placed in vibration end, frequency of oscillation depends on the quality of quartzy conical pipe characteristic and it.When sampling air flow is logical
Filter membrane, particulate matter therein are deposited on filter membrane, and filter membrane mass change causes frequency of oscillation to change, by measuring frequency of oscillation
Variation calculate the quality for being deposited on particulate matter on filter membrane, further according to sampling flow, sampling location environment temperature and barometer
Calculate the particulate matter standard state mass concentration of the period.
The advantages of oscillating balance method is that quantitative relationship is clear.But there are following two disadvantages for it: (1) volatilizing after sample heating
The loss of property and half volatile substance, causes measurement result relatively low;(2) need to install additional film dynamic measurement system (Filter
Dynamic Measurement System) relatively low result is calibrated.
β ray method is based on two hypothesis: the quartzy sampling membrane band of instrument is uniform;The PM2.5 particle collected
Physical characteristic is uniform, identical to β radiation attenuation rate.But under current conditions, two above hypothesis is generally difficult to set up, because
This determination data is generally acknowledged to that there is also deviations.The shortcomings that this method is in relative clean and dry regional failure rate
It is low, and it is very high in moist high-temperature area failure rate.
Summary of the invention
Present invention aim to address the measuring method of existing PM2.5, there are the complicated for operation and inaccurate skills of measurement result
Art problem provides a kind of recognition methods of PM2.5 pollution index based on characteristics of image.
In order to solve the above technical problems, the technical solution adopted by the present invention are as follows:
A kind of recognition methods of the PM2.5 pollution index based on characteristics of image, comprising the following steps:
Step 1. collects the history image for being labelled with PM2.5 measured value, and the history image of collection is all uniformly scaled
XS×YSPixel size;If image be it is colored, be just converted into grayscale image;
Step 2. calculates each scaling and is converted to the characteristics of image of image after grayscale image;
Step 3. obtains the prediction model of PM2.5 using homing method using the feature of each image as input;
Step 4. acquires target image, and target image is scaled X according to method same in step 1S×YSPixel is big
It is small, if image be it is colored, be converted into grayscale image;
Step 5. calculates according to the same method in step 2 for the target image after scaling and scales and be converted to ash
The characteristics of image of image after degree figure;
Step 6. is calculated using the characteristics of image of target image as the input of trained prediction model in step 3
Obtain the PM2.5 value of target image.
It is calculating each scaling in the step 2 and is being converted to after grayscale image after the characteristics of image of image, in addition it is also necessary to is right
The characteristics of image of each image and the value of PM2.5 are normalized;For the target image after scaling, In in the step 5
It calculates and scales and be converted to after grayscale image after the characteristics of image of image, it is also necessary to the characteristics of image of target image according to step
Same method for normalizing, is normalized in 2.In machine learning, using it is same trained when the same method data are carried out
Pretreatment, i.e., using it is same trained when the same method convert the image into gray value, calculate characteristics of image and be simultaneously normalized, In
Prediction Shi Caineng, which is carried out, using previous trained prediction model obtains optimal prediction result.
Use formula GRAY=0.299 × R+0.587 × G+0.114 × B by color image in the step 1 or step 4
Gray level image is converted to, wherein R, G, B respectively indicate the red of color image, green and blue wave band;
Characteristics of image in the step 2 or step 5 is the number N of the average pixel value of image, Local ExtremumMax, office
The mean μ of portion's varianceLVAR, image level direction and vertical direction second differnce weighted average, image level direction and vertical
Weighted average, the Q of the second order gradient sum in directionx、QyAnd QgThe weighted average of broken line difference and the spot number in image;
These features why are selected, are because finding that these features can be good at reflecting in actual photographed image during the experiment
PM2.5 height.
The average pixel value of described image is after the pixel value of all pixels adds up divided by XS×YS;
The number N of described image Local ExtremumMaxWith the mean μ of local varianceLVARCalculation method are as follows:
(1) N is setMaxAnd μLVARInitial value be 0;
(2) boundary pixel non-for each of image finds out 8 neighbouring pixels, remembers the composition set of 8 pixels
For PIXA;If current cell coordinate is (i, j), i and j respectively indicate the pixel and image most upper left corner pixel respectively vertical
With horizontal direction with the pixel number of current Pixel-space, then the coordinate of its 8 neighbours be respectively (i+1, j), (i, j+1),
(i-1, j), (i, j-1), (i-1, j-1), (i-1, j+1), (i+1, j-1) and (i+1, j+1);Here non-boundary pixel refers to
It is all in image to find above-mentioned 8 neighbours, and any neighbour's pixel pixel value is not empty pixel;
(3) if current pixel value is greater than 8 pixels on periphery, NMaxValue plus 1;And
Here, p indicates that current pixel, I (p) indicate the gray value of pixel p,Indicate PIXAIn 8 pixels and
The mean value of current pixel pixel value;
(4) after all pixels have all been handled, μLVAR=μLVAR/((XS-2)×(YS-2));
Described image horizontally and vertically second differnce weighted average, image level direction and vertical direction two
The calculation method of the weighted average of ladder degree sum are as follows:
(1) second differnce for calculating separately each pixel p of image level direction x and vertical direction y, is denoted as respectivelyWithAnd the second differnce of both direction is arranged according to sequence from small to large respectively
Sequence is removed repetition values in sequencer procedure, and is filtered out less than horizontal direction given threshold value αxOr vertical direction given threshold value αyInstitute
There is difference value, thus obtaining two sizes is respectively nxAnd nyHorizontal direction queue QxWith vertical direction queue Qy, two queues
In i-th element be respectivelyWithTwo scales corresponding to the pixel of i are as ordered as in queue
Score value;
(2) calculate each pixel p of image second order gradient andAnd by second order gradient and
Value removes repetition values in sequencer procedure according to sequence permutation from small to large, and filters out less than given threshold value αgAll two
Ladder degree and value, thus obtaining size is ngSecond order gradient and queue Qg, i-th element is respectively Q in queueg(i)=grad
(pi), second order gradient and value corresponding to the pixel of i are as ordered as in queue;
(3) respectively to three queue Qx、QyAnd QgCalculate corresponding broken line total lengthWithAnd overall weighting change degree With
(4) three queue Q are calculated separatelyx、QyAnd QgWeighted average WVLx=Vx/Lx、WVLy=Vy/LyAnd WVLg=
Vg/Lg, these three values just be respectively horizontal direction second differnce weighted average, vertical direction second differnce weighted average with
And the weighted average of second order gradient sum;
The calculation method of spot number in described image are as follows:
(1) a series of given ordinal scale value σ1,σ2,…,σs, s >=3, and some radius r is selected, it is drawn according to standardization
This Gauss operator of pula generates the convolution mask of corresponding each scale, wherein the centre coordinate of each convolution mask is (0,0), institute
There are all grids for differing r pixel with convolution mask central horizontal or vertical direction to belong to template;All nets in template
The coordinate set of lattice is { (tx,ty)|tx=-r ,-r+1 ..., r-1, r ∧ ty=-r ,-r+1 ..., r-1, r }, corresponding scale σ ∈
{σ1,σ2,…,σsConvolution mask in each grid (tx,ty) on value use following formula calculate:
To each scale σ ∈ { σ1,σ2,…,σs, convolution algorithm is done to image using corresponding convolution mask, obtains a system
Column convolved image I1,I2,…,Is;
(2) for each convolved image Ii, each pixel p in i=2 ..., s-1, it is assumed that its coordinate in the picture is
(i, j), the cell coordinates of 8 arest neighbors be respectively (i+1, j), (i, j+1), (i-1, j), (i, j-1), (i-1, j-1),
(i-1, j+1), (i+1, j-1) and (i+1, j+1);If IiIt (p) is the smallest in all neighbours, and Ii(p) also compare Ii-1With
Ii+1In all coordinates be (i, j), (i+1, j), (i, j+1), (i-1, j), (i, j-1), (i-1, j-1), (i-1, j+1), (i+
1, j-1) and the value of the pixel of (i+1, j+1) is all small, then it is exactly a spot that the pixel is corresponding;If IiIt (p) is all
It is maximum in neighbour, and Ii(p) also compare Ii-1And Ii+1In all coordinates be (i, j), (i+1, j), (i, j+1), (i-1, j),
The value of the pixel of (i, j-1), (i-1, j-1), (i-1, j+1), (i+1, j-1) and (i+1, j+1) is all big, then the pixel is corresponding
Be also a spot;In all convolved image Ii, i=2 ..., the dropout count found on s-1 is the spot in image
Number.
The Qx、QyAnd QgThe calculation method of the weighted average of broken line difference are as follows:
(1) first against Qx、QyAnd QgBroken line calculate separately the difference between two neighboring element, and it is this is poor
Value is respectively formed a queue, is denoted as QQ respectively according to sorting from small to largex、QQyAnd QQg, i-th of element is respectively QQx
(i)、QQy(i) and QQg(i)。
(2) respectively to three queue QQx、QQyAnd QQgCalculate corresponding broken line total lengthWithAnd overall weighting change degree With
(3) three queue QQ are calculated separatelyx、QQyAnd QQgWeighted average WVLQx=VQx/LQx、WVLQy=VQy/LQy
And WVLQg=VQg/LQg, these three values are just respectively Qx、QyAnd QgThe weighted average of broken line difference.
The feature of image and the value of PM2.5 are normalized according to following formula in the step 2 and step 5:
Wherein F indicates some feature, and F (I) is indicated to scale and is converted to history image or target image I after grayscale image
Value on feature F, minFFeature F is indicated in all scalings and is converted to the minimum value on the history image after grayscale image,
maxFFeature F is indicated in all scalings and is converted to the maximum value on the history image after grayscale image, Fnorm(I) indicate I in feature
Value after being normalized on F, and if to being converted to F when the target image after grayscale image is normalizednormIt (I) will when > 1
Fnorm(I) it is set to 1, if Fnorm(I) by F when < 0norm(I) it is set to 0, InewFor to normalized new value.This method implements
It is fairly simple efficient, therefore this method is selected to be normalized.
The method for obtaining the prediction model of PM2.5 in the step 3 using homing method is feedforward neural network, Gauss mistake
The methods of journey or common least square.
The concrete operation step of the prediction model that PM2.5 is obtained using feedforward neural network method are as follows:
(1) activation primitive is set in feedforward neural network, this several or hyperbolic of writing of logic can be used in this step
Tangent function;
(2) level and every layer of neuronal quantity of neural network are set;
(3) back-propagation algorithm learning neural network is used;
(4) network learnt is preserved, need to save in have and used which kind of activation primitive, neural network
The number of plies, the weight of every layer of neuronal quantity and each neuron;
When obtaining the prediction model of PM2.5 using feedforward neural network method in the step 6, need to read in the base of preservation
In the prediction model parameters of feedforward neural network, including activation primitive, the number of plies of neural network, every layer of neuronal quantity and
The weight of each neuron, then according to the feedforward transmission method of feedforward neural network using the characteristics of image after normalization as defeated
Enter to calculate InewPM2.5 value PM2.5 after corresponding normalizationnorm(Inew), finally by it according to the anti-normalizing of following formula
PM2.5 value PM2.5 (I after change as new shooting imagenew):
PM2.5(Inew)=PM2.5norm(Inew)×(maxF-minF)+minF。
The present invention makes full use of the measured value and corresponding outdoor scene photo of history PM2.5, comprehensive to use image processing techniques
And data mining technology, it is only necessary to can quickly estimate its PM2.5 index using target area scenery photo, overcome tradition
Method is typically only capable to the shortcomings that one-point measurement, provides a kind of technical method that can quickly estimate extensive area PM2.5.This
Invention use above technical scheme, compared with prior art the advantages of be:
(1) it does not need to make or uses special instrument and chemical reagent, using only the photo energy of target area
Enough identify PM2.5 index;
(2) it when only target scene photograph, can not rebuild or the environmental factor of repeated observation target area, this method
Still its PM2.5 index can be identified;
(3) it is capable of the average PM2.5 value of Direct Recognition large area region, rather than ocean weather station observation.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention;
Fig. 2-5 is four input pictures in present example.
Specific embodiment
The recognition methods of PM2.5 pollution index of one of the present embodiment based on characteristics of image, comprising the following steps:
Step 1. selects a region, shoots the photochrome of different time using camera and records this time
PM2.5 value is labelled with the history image of PM2.5 measured value to collect, and all photos are divided into daytime according to season and time
With evening two major classes, calculated all by taking daytime as an example below;The history image of collection is all uniformly scaled XS×YSPixel is big
Small, used here as the convert order of imagemagick software: " convert-resize 640x320 " is by the history of collection
Image is all uniformly scaled 640 × 320 pixel sizes;It will scheme used here as the mode for the red band for only using photochrome
As being converted to grayscale image, gray scale can also be converted the image into using formula GRAY=0.299 × R+0.587 × G+0.114 × B
Figure, wherein R, G, B respectively indicate the red of color image, green and blue wave band;
Step 2. calculates each scaling and is converted to the number of the average pixel value of image after grayscale image, Local Extremum
NMax, local variance mean μLVAR, image level direction and vertical direction second differnce weighted average, image level direction
With weighted average, the Q of the second order gradient sum of vertical directionx、QyAnd QgThe weighted average and the spot in image of broken line difference
Points;
The average pixel value of described image is after the pixel value of all pixels adds up divided by XS×YS;
The number N of described image Local ExtremumMaxWith the mean μ of local varianceLVARCalculation method are as follows:
(1) N is setMaxAnd μLVARInitial value be 0;
(2) boundary pixel non-for each of image finds out 8 neighbouring pixels, remembers the composition set of 8 pixels
For PIXA;If current cell coordinate is (i, j), i and j respectively indicate the pixel and image most upper left corner pixel respectively vertical
With horizontal direction with the pixel number of current Pixel-space, then the coordinate of its 8 neighbours be respectively (i+1, j), (i, j+1),
(i-1, j), (i, j-1), (i-1, j-1), (i-1, j+1), (i+1, j-1) and (i+1, j+1);Here non-boundary pixel refers to
It is all in image to find above-mentioned 8 neighbours, and any neighbour's pixel pixel value is not empty pixel;
(3) if current pixel value is greater than 8 pixels on periphery, NMaxValue plus 1;And
Here, p indicates that current pixel, I (p) indicate the gray value of pixel p,Indicate PIXAIn 8 pixels and
The mean value of current pixel pixel value;
(4) after all pixels have all been handled, μLVAR=μLVAR/((XS-2)×(YS-2));
Described image horizontally and vertically second differnce weighted average, image level direction and vertical direction two
The calculation method of the weighted average of ladder degree sum are as follows:
(1) second differnce for calculating separately each pixel p of image level direction x and vertical direction y, is denoted as respectivelyWithAnd the second differnce of both direction is arranged according to sequence from small to large respectively
Sequence is removed repetition values in sequencer procedure, and is filtered out less than horizontal direction given threshold value αxOr vertical direction given threshold value αyInstitute
There is difference value, thus obtaining two sizes is respectively nxAnd nyHorizontal direction queue QxWith vertical direction queue Qy, two queues
In i-th element be respectivelyWithSecond order corresponding to the pixel of i is as ordered as in queue
Difference value;
(2) calculate each pixel p of image second order gradient andAnd by second order gradient and
Value removes repetition values in sequencer procedure according to sequence permutation from small to large, and filters out less than given threshold value αgAll two
Ladder degree and value, thus obtaining size is ngSecond order gradient and queue Qg, i-th element is respectively Q in queueg(i)=grad
(pi), second order gradient and value corresponding to the pixel of i are as ordered as in queue;
(3) respectively to three queue Qx、QyAnd QgCalculate corresponding broken line total lengthWithAnd overall weighting change degree With
(4) three queue Q are calculated separatelyx、QyAnd QgWeighted average WVLx=Vx/Lx、WVLy=Vy/LyAnd WVLg=
Vg/Lg, these three values just be respectively horizontal direction second differnce weighted average, vertical direction second differnce weighted average with
And the weighted average of second order gradient sum;
The Qx、QyAnd QgThe calculation method of the weighted average of broken line difference are as follows:
(1) first against Qx、QyAnd QgBroken line calculate separately the difference between two neighboring element, and it is this is poor
Value is respectively formed a queue, is denoted as QQ respectively according to sorting from small to largex、QQyAnd QQg, i-th of element is respectively QQx
(i)、QQy(i) and QQg(i)。
(2) respectively to three queue QQx、QQyAnd QQgCalculate corresponding broken line total lengthWithAnd overall weighting change degree With
(3) three queue QQ are calculated separatelyx、QQyAnd QQgWeighted average WVLQx=VQx/LQx、WVLQy=VQy/LQy
And WVLQg=VQg/LQg, these three values are just respectively Qx、QyAnd QgThe weighted average of broken line difference.
The calculation method of spot number in described image are as follows:
(1) a series of given ordinal scale value σ1,σ2,…,σs, s >=3, and some radius r is selected, it is drawn according to standardization
This Gauss operator of pula generates the convolution mask of corresponding each scale, wherein the centre coordinate of each convolution mask is (0,0), institute
There are all grids for differing r pixel with convolution mask central horizontal or vertical direction to belong to template;All nets in template
The coordinate set of lattice is { (tx, ty)|tx=-r ,-r+1 ..., r-1, r ∧ ty=-r ,-r+1 ..., r-1, r }, corresponding scale σ ∈
{σ1,σ2,…,σsConvolution mask in each grid (tx, ty) on value use following formula calculate:
To each scale σ ∈ { σ1,σ2,…,σs, convolution algorithm is done to image using corresponding convolution mask, obtains a system
Column convolved image I1,I2,…,Is;
(2) for each convolved image Ii, each pixel p in i=2 ..., s-1, it is assumed that its coordinate in the picture is
(i, j), the cell coordinates of 8 arest neighbors be respectively (i+1, j), (i, j+1), (i-1, j), (i, j-1), (i-1, j-1),
(i-1, j+1), (i+1, j-1) and (i+1, j+1);If IiIt (p) is the smallest in all neighbours, and Ii(p) also compare Ii-1With
Ii+1In all coordinates be (i, j), (i+1, j), (i, j+1), (i-1, j), (i, j-1), (i-1, j-1), (i-1, j+1), (i+
1, j-1) and the value of the pixel of (i+1, j+1) is all small, then it is exactly a spot that the pixel is corresponding;If IiIt (p) is all
It is maximum in neighbour, and Ii(p) also compare Ii-1And Ii+1In all coordinates be (i, j), (i+1, j), (i, j+1), (i-1, j),
The value of the pixel of (i, j-1), (i-1, j-1), (i-1, j+1), (i+1, j-1) and (i+1, j+1) is all big, then the pixel is corresponding
Be also a spot;In all convolved image Ii, i=2 ..., the dropout count found on s-1 is the spot in image
Number;
This example has selected to calculate the number N of Local ExtremumMax, local variance mean μLVAR, average pixel value, base
In the second order gradient and weighted average and Q of Laplace operatorgThe weighted average of broken line difference, the Q based on Sobel operatory
Spot number these features in the weighted average of broken line difference and image based on Gauss-Laplace, it is calculated
Image section feature is as shown in the table:
The library OpenCV is used when calculating Laplacian Image Speckle number, partial code is as follows:
Mat image=img;
Mat descriptors;
vector<KeyPoint>keypoints;
SimpleBlobDetector::Params params;
Params.minThreshold=10;
Params.maxThreshold=100;
Params.thresholdStep=10;
Params.minArea=10;
Params.minConvexity=0.3;
Params.minInertiaRatio=0.01;
Params.maxArea=8000;
Params.maxConvexity=10;
Params.filterByColor=false;
Params.filterByCircularity=false;
SimpleBlobDetector blobDetector(params);
blobDetector.create("SimpleBlob");
blobDetector.detect(image,keypoints);
drawKeypoints(image,keypoints,image,Scalar(255,0,0));
cout<<keypoints.size()<<"";
Second order gradient and weighted average and Q of the calculating based on Laplace operatorgThe weighted average of broken line difference
The library OpenCV is used, partial code is as follows:
Here αgThreshold value is set as 50.
And the value of each characteristics of image and PM2.5 to each image is normalized according to following formula:
Wherein F indicates some feature, and F (I) is indicated to scale and is converted to history image or target image I after grayscale image
Value on feature F, minFFeature F is indicated in all scalings and is converted to the minimum value on the history image after grayscale image,
maxFFeature F is indicated in all scalings and is converted to the maximum value on the history image after grayscale image, Fnorm(I) indicate I in feature
Value after being normalized on F, and if to being converted to F when the target image after grayscale image is normalizednormIt (I) will when > 1
Fnorm(I) it is set to 1, if Fnorm(I) by F when < 0norm(I) it is set to 0, InewFor to normalized new value.It is with average pixel value
, the maximum value of this feature is 147.323 in training data, minimum value 49.6862, therefore, when this value is 113.21
When, the value after normalization is (113.21-49.6862)/(147.323-49.6862)=0.181324.Following table is each spy
Result after the PM2.5 that seeks peace normalization:
Step 3. obtains the prediction mould of PM2.5 using feedforward neural network method using the feature of each image as input
Type, concrete operation step are as follows:
(1) activation primitive is set in feedforward neural network, this several or hyperbolic of writing of logic can be used in this step
Tangent function, it is this number of writing of logic, also referred to as sigmoid function that activation primitive is arranged here;
(2) level and every layer of neuronal quantity of neural network are set, it is 3 that the network number of plies is arranged here, wherein inputting
Layer neuron number is 7, and output layer neuron number is 1, and middle layer neuron number is 1000;
(3) back-propagation algorithm learning neural network is used, the minimum trained rate of parameter used herein is 0.001, most
Height training rate is 0.01, and burst error is [0.001,0.02], the number of iterations 10000;
Neural network uses the library of 0.998 version of lwneuralnetplus, and part training code is as follows:
Iomanager=new iomanagelwnnfann ();
iomanager->info_from_file(argv[1],&npatterns,&ninput,&noutput);
Net=new network (network::LOGISTIC, 3, ninput, 1000, noutput);
net->set_momentum(0);
net->set_learning_rate(0.001);
net->jolt(0.03,0.22);
Train=new trainer (net, " ", " ");
train->set_iomanager(iomanager);
cout<<"Loading training data..."<<endl;
train->load_training(argv[1]);
train_go(argv,train,net);
Wherein train_go partial code is as follows:
(4) network learnt is preserved, need to save in have and used which kind of activation primitive, neural network
The number of plies, the weight of every layer of neuronal quantity and each neuron;
Step 4. acquires target image, and target image is scaled X according to method same in step 1S×YSPixel is big
It is small, if image be it is colored, be converted into grayscale image;
Step 5. calculates according to the same method in step 2 for the target image after scaling and scales and be converted to ash
Spend the number N of the average pixel value of image after figure, Local ExtremumMax, local variance mean μLVAR, image level direction and
Weighted average, the Q of the second order gradient sum of vertical direction second differnce weighted average, image level direction and vertical directionx、
QyAnd QgThe weighted average of broken line difference and the spot number in image, and to each characteristics of image of target image according to step 2
In same method for normalizing, be normalized;
Step 6. is read in using the characteristics of image of target image as the input of trained prediction model in step 3
The prediction model parameters based on feedforward neural network saved, including activation primitive, the number of plies of neural network, every layer of neuron
The weight of quantity and each neuron, it is then according to the feedforward transmission method of feedforward neural network that the image after normalization is special
Sign calculates I as inputnewPM2.5 value PM2.5 after corresponding normalizationnorm(Inew), finally by it according to following public affairs
PM2.5 value PM2.5 (I after formula renormalization as new shooting imagenew):
PM2.5(Inew)=PM2.5norm(Inew)×(maxF-minF)+minF。
Four images that we have selected in the following figure are identified as new shooting image.It needs exist for first in the following figure
The feature of each image is normalized, the PM2.5 value after then bringing previously trained neural computing normalization into, most
Renormalization is carried out afterwards.Recognition effect is as shown in the table:
Measured value | Predicted value | |
Image 1 | 37 | 49 |
Image 2 | 223 | 247 |
Image 3 | 323 | 306 |
Image 4 | 117 | 103 |
No matter using which kind of method for normalizing in the step 2 and step 5, as long as can be by characteristic value or PM2.5 value
The number for becoming [0,1] section or (0,1) section, just belongs to patent claims range;
The optional method for building up of prediction model in the step 3 include neural network, Gaussian process and it is common most
Small two the methods of multiply etc., no matter using which kind of method, as long as used during prediction neural network, Gaussian process and
One or more of the methods of common least square feature is all in the scope of the claims of this patent;
In the step 2 and step 5, image level direction and vertical direction second differnce weighted average and two are calculated
A variety of methods can be used in the weighted average of ladder degree sum, compute repeatedly and obtain various features, such as use Sobel respectively
Operator and Laplace operator calculate these three features;
The step 2 and step 5 are normalized to optional step, if not using step 2 and step 5, need to walk again
The PM2.5 value predicted in rapid 6 is without carrying out renormalization operation.
The value of prediction target PM2.5 in this patent may be visibility etc. can be used that real number measured other
Actual prediction target, no matter using which kind of prediction target, as long as image level direction and vertical direction second differnce has been used to add
The weighted average and Q of weight average value, second order gradient sumx、QyAnd QgThe weighted average of broken line difference these features are just all
Belong to the scope of the claims of this patent.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should be included within scope of the presently claimed invention.
Claims (6)
1. a kind of recognition methods of the PM2.5 pollution index based on characteristics of image, it is characterised in that: the following steps are included:
Step 1. collects the history image for being labelled with PM2.5 measured value, and the history image of collection is all uniformly scaled XS×YS
Pixel size;If image be it is colored, be just converted into grayscale image;
Step 2. calculates each scaling and is converted to the characteristics of image of image after grayscale image;
Step 3. obtains the prediction model of PM2.5 using homing method using the feature of each image as input;
Step 4. acquires target image, and target image is scaled X according to method same in step 1S×YSPixel size, such as
Fruit image be it is colored, be converted into grayscale image;
Step 5. calculates according to the same method in step 2 for the target image after scaling and scales and be converted to grayscale image
The characteristics of image of image afterwards;
Step 6. is calculated using the characteristics of image of target image as the input of trained prediction model in step 3
The PM2.5 value of target image;
It is calculating each scaling in the step 2 and is being converted to after grayscale image after the characteristics of image of image, in addition it is also necessary to each
The characteristics of image of image and the value of PM2.5 are normalized;For the target image after scaling in the step 5, calculating
It scales and is converted to after grayscale image after the characteristics of image of image, it is also necessary to the characteristics of image of target image according in step 2
Same method for normalizing, is normalized;
Color image is converted using formula GRAY=0.299 × R+0.587 × G+0.114 × B in the step 1 or step 4
For gray level image, wherein R, G, B respectively indicate the red of color image, green and blue wave band;
Characteristics of image in the step 2 or step 5 is the number N of the average pixel value of image, Local ExtremumMax, part side
The mean μ of differenceLVAR, image level direction and vertical direction second differnce weighted average, image level direction and vertical direction
Second order gradient sum weighted average, Qx、QyAnd QgThe weighted average of broken line difference and the spot number in image, it is described
Qx、Qy、QgIt is defined as follows and states described in step a-b;
The average pixel value of described image is after the pixel value of all pixels adds up divided by XS×YS;
The number N of the Local Extremum of described imageMaxWith the mean μ of local varianceLVARCalculation method are as follows:
(1) N is setMaxAnd μLVARInitial value be 0;
(2) boundary pixel non-for each of image finds out 8 neighbouring pixels, remembers that the composition collection of 8 pixels is combined into
PIXA;If current cell coordinate is (i, j), i and j respectively indicate the pixel and image most upper left corner pixel respectively vertical and
Horizontal direction with current Pixel-space pixel number, then the coordinate of its 8 neighbours be respectively (i+1, j), (i, j+1), (i-
1, j), (i, j-1), (i-1, j-1), (i-1, j+1), (i+1, j-1) and (i+1, j+1);Here non-boundary pixel refers to figure
It is all as in find above-mentioned 8 neighbours, and any neighbour's pixel pixel value is not empty pixel;
(3) if current pixel value is greater than 8 pixels on periphery, NMaxValue plus 1;And
Here, p indicates that current pixel, I (p) indicate the gray value of pixel p,Indicate PIXAIn 8 pixels and current
The mean value of pixel pixel value;
(4) after all pixels have all been handled, μLVAR=μLVAR/((XS-2)×(YS-2));
Described image horizontally and vertically two ladder of second differnce weighted average, image level direction and vertical direction
Spend the calculation method of the weighted average of sum are as follows:
A. the second differnce for calculating separately each pixel p of image level direction x and vertical direction y, is denoted as respectivelyWithAnd the second differnce of both direction is arranged according to sequence from small to large respectively
Sequence is removed repetition values in sequencer procedure, and is filtered out less than horizontal direction given threshold value αxOr vertical direction given threshold value αyInstitute
There is difference value, thus obtaining two sizes is respectively nxAnd nyHorizontal direction queue QxWith vertical direction queue Qy, two queues
In i-th element be respectivelyWithTwo scales corresponding to the pixel of i are as ordered as in queue
Score value;
B. calculate each pixel p of image second order gradient andAnd by second order gradient and value according to
Sequence permutation from small to large removes repetition values in sequencer procedure, and filters out less than given threshold value αgAll second order gradients
And value, thus obtaining size is ngSecond order gradient and queue Qg, i-th element is respectively Q in queueg(i)=grad (pi),
Second order gradient and value corresponding to the pixel of i are as ordered as in queue;
C. respectively to three queue Qx、QyAnd QgCalculate corresponding broken line total lengthWith
And overall weighting change degree With
D. three queue Q are calculated separatelyx、QyAnd QgWeighted average WVLx=Vx/Lx、WVLy=Vy/LyAnd WVLg=Vg/Lg,
These three values are just respectively horizontal direction second differnce weighted average, vertical direction second differnce weighted average and second order
The weighted average of gradient sum;
The calculation method of spot number in described image are as follows:
(1) a series of given ordinal scale value σ1,σ2,…,σs, s >=3, and some radius r is selected, according to standardization La Pula
This Gauss operator generates the convolution mask of corresponding each scale, wherein the centre coordinate of each convolution mask is (0,0), Suo Youhe
All grids that convolution mask central horizontal or vertical direction differ r pixel belong to template;All grids in template
Coordinate set is { (tx,ty)|tx=-r ,-r+1 ..., r-1, r ∧ ty=-r ,-r+1 ..., r-1, r }, corresponding scale σ ∈ { σ1,
σ2,…,σsConvolution mask in each grid (tx,ty) on value use following formula calculate:
To each scale σ ∈ { σ1,σ2,…,σs, convolution algorithm is done to image using corresponding convolution mask, obtains a series of volumes
Product image I1,I2,…,Is;
(2) for each convolved image Ii, each pixel p in i=2 ..., s-1, it is assumed that its coordinate in the picture be (i,
J), the cell coordinate of 8 arest neighbors be respectively (i+1, j), (i, j+1), (i-1, j), (i, j-1), (i-1, j-1), (i-1,
J+1), (i+1, j-1) and (i+1, j+1);If IiIt (p) is the smallest in all neighbours, and Ii(p) also compare Ii-1And Ii+1In
All coordinates are (i, j), (i+1, j), (i, j+1), (i-1, j), (i, j-1), (i-1, j-1), (i-1, j+1), (i+1, j-1)
The value of the pixel of (i+1, j+1) is all small, then it is exactly a spot that the pixel is corresponding;If IiIt (p) is in all neighbours
It is maximum, and Ii(p) also compare Ii-1And Ii+1In all coordinates be (i, j), (i+1, j), (i, j+1), (i-1, j), (i, j-
1), the value of the pixel of (i-1, j-1), (i-1, j+1), (i+1, j-1) and (i+1, j+1) is all big, then the pixel is corresponding
It is a spot;In all convolved image Ii, i=2 ..., the dropout count found on s-1 is the spot number in image.
2. a kind of recognition methods of PM2.5 pollution index based on characteristics of image according to claim 1, feature exist
In: the Qx、QyAnd QgThe calculation method of the weighted average of broken line difference are as follows:
(1) first against Qx、QyAnd QgBroken line calculate separately the difference between two neighboring element, and this difference is pressed
According to sorting from small to large, it is respectively formed a queue, is denoted as QQ respectivelyx、QQyAnd QQg, i-th of element is respectively QQx(i)、
QQy(i) and QQg(i);
(2) respectively to three queue QQx、QQyAnd QQgCalculate corresponding broken line total lengthWithAnd overall weighting change degree With
(3) three queue QQ are calculated separatelyx、QQyAnd QQgWeighted average WVLQx=VQx/LQx、WVLQy=VQy/LQyWith
WVLQg=VQg/LQg, these three values are just respectively Qx、QyAnd QgThe weighted average of broken line difference.
3. a kind of recognition methods of PM2.5 pollution index based on characteristics of image according to claim 2, feature exist
In:
The feature of image and the value of PM2.5 are normalized according to following formula in the step 2 and step 5:
Wherein F indicates some feature, and F (I) indicates to scale and be converted to history image after grayscale image or target image I in spy
Levy the value on F, minFFeature F is indicated in all scalings and is converted to the minimum value on the history image after grayscale image, maxFTable
Show feature F in all scalings and is converted to the maximum value on the history image after grayscale image, Fnorm(I) indicate that I returns on feature F
One change after value, and if to being converted to F when the target image after grayscale image is normalizednorm(I) by F when > 1norm(I)
It is set to 1, if Fnorm(I) by F when < 0norm(I) it is set to 0, InewFor to normalized new value.
4. a kind of recognition methods of PM2.5 pollution index based on characteristics of image according to claim 3, feature exist
In: obtained in the step 3 using homing method the prediction model of PM2.5 method be feedforward neural network, Gaussian process or
Common least square method.
5. a kind of recognition methods of PM2.5 pollution index based on characteristics of image according to claim 4, feature exist
In: the method for the prediction model that PM2.5 is obtained using homing method is feedforward neural network, concrete operation step are as follows:
(1) activation primitive is set in feedforward neural network, this several or tanh of writing of logic can be used in this step
Function;
(2) level and every layer of neuronal quantity of neural network are set;
(3) back-propagation algorithm learning neural network is used;
(4) network learnt is preserved, need to save in have which kind of activation primitive used, the layer of neural network
Several, every layer of neuronal quantity and the weight of each neuron.
6. a kind of recognition methods of PM2.5 pollution index based on characteristics of image according to claim 5, feature exist
In: when obtaining the prediction model of PM2.5 using feedforward neural network method, need to read in preservation based on feedforward neural network
Prediction model parameters, including activation primitive, the power of the number of plies of neural network, every layer of neuronal quantity and each neuron
Then weight calculates I for the characteristics of image after normalization as input according to the feedforward transmission method of feedforward neural networknewInstitute
PM2.5 value PM2.5 after corresponding normalizationnorm(Inew), finally it is shot according to after following formula renormalization as new
The PM2.5 value PM2.5 (I of imagenew):
PM2.5(Inew)=PM2.5norm(Inew)×(maxF-minF)+minF。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710301867.4A CN107044947B (en) | 2017-05-02 | 2017-05-02 | A kind of recognition methods of the PM2.5 pollution index based on characteristics of image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710301867.4A CN107044947B (en) | 2017-05-02 | 2017-05-02 | A kind of recognition methods of the PM2.5 pollution index based on characteristics of image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107044947A CN107044947A (en) | 2017-08-15 |
CN107044947B true CN107044947B (en) | 2019-11-19 |
Family
ID=59546232
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710301867.4A Active CN107044947B (en) | 2017-05-02 | 2017-05-02 | A kind of recognition methods of the PM2.5 pollution index based on characteristics of image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107044947B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109087277B (en) * | 2018-06-11 | 2021-02-26 | 北京工业大学 | Method for measuring PM2.5 of fine air particles |
CN111488804B (en) * | 2020-03-19 | 2022-11-11 | 山西大学 | Labor insurance product wearing condition detection and identity identification method based on deep learning |
CN111912755B (en) * | 2020-08-07 | 2021-08-10 | 山东中煤工矿物资集团有限公司 | Mining dust concentration sensor, sensor system and method |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6310346B1 (en) * | 1997-05-30 | 2001-10-30 | University Of Central Florida | Wavelength-tunable coupled antenna uncooled infrared (IR) sensor |
CN103903273A (en) * | 2014-04-17 | 2014-07-02 | 北京邮电大学 | PM2.5 grade fast-evaluating system based on mobile phone terminal |
CN103954542A (en) * | 2014-05-12 | 2014-07-30 | 中国计量学院 | PM2.5 (Particulate Matter2.5) concentration detector based on definition evaluation without reference image |
CN104462778A (en) * | 2014-11-06 | 2015-03-25 | 华北电力大学 | PM2.5 pollutant measurement method based on deep learning |
CN103593660B (en) * | 2013-11-27 | 2016-08-17 | 青岛大学 | The palm grain identification method that gradient of intersecting under a kind of invariant feature image encodes |
CN106295516A (en) * | 2016-07-25 | 2017-01-04 | 天津大学 | Haze PM2.5 value method of estimation based on image |
CN106401359A (en) * | 2016-08-31 | 2017-02-15 | 余姚市泗门印刷厂 | Infrared-photographing-based window control platform |
-
2017
- 2017-05-02 CN CN201710301867.4A patent/CN107044947B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6310346B1 (en) * | 1997-05-30 | 2001-10-30 | University Of Central Florida | Wavelength-tunable coupled antenna uncooled infrared (IR) sensor |
CN103593660B (en) * | 2013-11-27 | 2016-08-17 | 青岛大学 | The palm grain identification method that gradient of intersecting under a kind of invariant feature image encodes |
CN103903273A (en) * | 2014-04-17 | 2014-07-02 | 北京邮电大学 | PM2.5 grade fast-evaluating system based on mobile phone terminal |
CN103954542A (en) * | 2014-05-12 | 2014-07-30 | 中国计量学院 | PM2.5 (Particulate Matter2.5) concentration detector based on definition evaluation without reference image |
CN104462778A (en) * | 2014-11-06 | 2015-03-25 | 华北电力大学 | PM2.5 pollutant measurement method based on deep learning |
CN106295516A (en) * | 2016-07-25 | 2017-01-04 | 天津大学 | Haze PM2.5 value method of estimation based on image |
CN106401359A (en) * | 2016-08-31 | 2017-02-15 | 余姚市泗门印刷厂 | Infrared-photographing-based window control platform |
Also Published As
Publication number | Publication date |
---|---|
CN107044947A (en) | 2017-08-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP3614308A1 (en) | Joint deep learning for land cover and land use classification | |
CN109800631A (en) | Fluorescence-encoded micro-beads image detecting method based on masked areas convolutional neural networks | |
CN107909109B (en) | SAR image classification method based on conspicuousness and multiple dimensioned depth network model | |
CN108921203B (en) | Detection and identification method for pointer type water meter | |
CN108334847A (en) | A kind of face identification method based on deep learning under real scene | |
CN104202547B (en) | Method, projection interactive approach and its system of target object are extracted in projected picture | |
CN108615046A (en) | A kind of stored-grain pests detection recognition methods and device | |
CN107665492A (en) | Colon and rectum panorama numeral pathological image tissue segmentation methods based on depth network | |
CN106682697A (en) | End-to-end object detection method based on convolutional neural network | |
CN107044947B (en) | A kind of recognition methods of the PM2.5 pollution index based on characteristics of image | |
CN109064462A (en) | A kind of detection method of surface flaw of steel rail based on deep learning | |
CN108629369A (en) | A kind of Visible Urine Sediment Components automatic identifying method based on Trimmed SSD | |
CN110517311A (en) | Pest and disease monitoring method based on leaf spot lesion area | |
CN109509187A (en) | A kind of efficient check algorithm for the nibs in big resolution ratio cloth image | |
CN104239902B (en) | Hyperspectral image classification method based on non local similitude and sparse coding | |
CN110569747A (en) | method for rapidly counting rice ears of paddy field rice by using image pyramid and fast-RCNN | |
CN111724355B (en) | Image measuring method for abalone body type parameters | |
CN107978110A (en) | Fence intelligence identifying system in place and recognition methods based on images match | |
CN109815953A (en) | One kind being based on vehicle annual test target vehicle identification matching system | |
CN113435282A (en) | Unmanned aerial vehicle image ear recognition method based on deep learning | |
CN111860459A (en) | Gramineous plant leaf stomata index measuring method based on microscopic image | |
CN111426637A (en) | Method for acquiring water quality parameters of urban black and odorous rivers by using unmanned aerial vehicle | |
KR101933856B1 (en) | System for image processing using the convolutional neural network and method using the same | |
CN110298410A (en) | Weak target detection method and device in soft image based on deep learning | |
CN112489026A (en) | Asphalt pavement disease detection method based on multi-branch parallel convolution neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20211222 Address after: 101100 4th floor, building 6, No.10 yard, Jiachuang Road, Tongzhou District, Beijing Patentee after: BEIJING JIAHUA ZHILIAN TECHNOLOGY CO.,LTD. Address before: 030006 No. 92, Hollywood Road, Taiyuan, Shanxi Patentee before: SHANXI University |
|
TR01 | Transfer of patent right |